Panel Johansen Cointegration Test
Also known as: panel Johansen test, Larsson-Lyhagen-Lothgren test, LLL panel cointegration, panel trace test
The Panel Johansen cointegration test extends Johansen's maximum-likelihood framework to panel data, allowing researchers to test whether multiple non-stationary variables share long-run equilibrium relationships across cross-sectional units. It pools the likelihood-ratio statistics from individual Johansen tests and compares the standardised average against a standard normal distribution, yielding greater power than single-country approaches.
Key highlights
- Detects multiple cointegrating vectors, unlike residual-based single-equation tests.
- Pooling across N cross-sections dramatically increases statistical power relative to individual-country Johansen tests.
- Allows heterogeneous short-run dynamics and lag orders across units while maintaining a common long-run framework.
- The standardised panel statistic has a simple standard normal asymptotic distribution, making inference straightforward.
- Compatible with subsequent VECM estimation to model both short-run adjustment and long-run equilibrium.
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 Panel Johansen cointegration when you have a balanced or near-balanced panel of multiple I(1) variables observed over a moderate to long time dimension (T ≥ 40 recommended) and you need to detect one or more long-run equilibrium relationships across cross-sectional units. It is particularly valuable in macroeconomic and finance panels — e.g., testing purchasing power parity, energy-growth nexus, or financial integration across countries — where individual-country time series are too short for the single-country Johansen test. Do not use it when variables are I(0) (already stationary), when T is very short (T < 20), or when the panel is severely unbalanced, as the asymptotic normal approximation breaks down. For a single cointegrating equation without multiple vectors, panel Engle-Granger or ARDL bounds tests are simpler alternatives.
Strengths & limitations
- Detects multiple cointegrating vectors, unlike residual-based single-equation tests.
- Pooling across N cross-sections dramatically increases statistical power relative to individual-country Johansen tests.
- Allows heterogeneous short-run dynamics and lag orders across units while maintaining a common long-run framework.
- The standardised panel statistic has a simple standard normal asymptotic distribution, making inference straightforward.
- Compatible with subsequent VECM estimation to model both short-run adjustment and long-run equilibrium.
- Requires a reasonably long time dimension (T ≥ 40) for each unit; small T causes the individual Johansen statistics to be unreliable.
- The standard Larsson-Lyhagen-Lothgren version assumes cross-sectional independence; cross-sectional dependence — common in macro panels — inflates the test statistic and requires bootstrap or CD-robust extensions.
- Choosing VAR lag length unit by unit is computationally intensive and sensitive to model specification.
- Assumes the cointegration rank is homogeneous across all units, which may be unrealistic in heterogeneous panels.
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
What is the difference between Panel Johansen and Panel Engle-Granger cointegration tests?
Panel Engle-Granger tests a single cointegrating relationship by running an OLS regression and testing the residuals for stationarity; it can miss additional cointegrating vectors. Panel Johansen uses a VAR framework that simultaneously identifies all cointegrating vectors, making it more powerful and informative when more than one long-run relationship may exist.
How many time periods do I need for the Panel Johansen test?
A minimum of T ≈ 40 periods per unit is commonly recommended, because each unit's Johansen trace statistic requires a well-behaved VAR. With T < 20 the individual statistics are unreliable and the normal approximation for the panel statistic may be poor. Increasing N compensates somewhat, but a short T is the binding constraint.
How do I handle cross-sectional dependence in the panel?
The Larsson-Lyhagen-Lothgren (2001) statistic assumes cross-sectional independence. If a Pesaran CD test or similar rejects independence, use a bootstrap version of the panel Johansen test or a factor-augmented extension that accounts for common factors driving co-movement across units.
What do I do after finding cointegration?
Estimate a Panel Vector Error Correction Model (Panel VECM). The VECM decomposes each variable's change into a long-run error-correction term (how fast the variable returns to equilibrium after a shock) and short-run dynamics, providing both the long-run cointegrating coefficients and the speed of adjustment.
Can the cointegration rank differ across panel units?
The standard Larsson-Lyhagen-Lothgren test imposes a common rank. If ranks are expected to differ — e.g., some countries are integrated while others are not — consider running individual Johansen tests per unit and reporting unit-specific results alongside the panel statistic, or use rank heterogeneity extensions in the literature.
Sources
- 1.Larsson, R., Lyhagen, J., & Lothgren, M. (2001). Likelihood-based cointegration tests in heterogeneous panels. Econometrics Journal, 4(1), 109–142.
- 2.Johansen, S. (1991). Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica, 59(6), 1551–1580.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Panel Johansen Cointegration. ScholarGate. https://scholargate.app/econometrics/panel-johansen-cointegration