Fourier Johansen Cointegration Test
Also known as: Fourier Johansen test, Fourier-Johansen trace test, smooth-break Johansen cointegration, FJ cointegration
The Fourier Johansen cointegration test extends the classical Johansen trace and maximum-eigenvalue tests by embedding low-frequency Fourier terms in the deterministic component of the VECM. This allows the test to remain valid when cointegrating relationships experience gradual, smooth regime shifts that standard Johansen critical values do not accommodate.
Key highlights
- Detects cointegration that standard Johansen misses when the relationship experiences smooth, gradual structural change.
- Does not require the researcher to pre-specify a break date or break type — the Fourier terms are estimated from the data.
- Extends naturally to multiple cointegrating vectors (rank > 1), preserving the full power of the multivariate Johansen framework.
- Low-frequency Fourier terms avoid overfitting; a single sine-cosine pair often captures the dominant smooth shift.
- Applicable to common empirical settings such as energy markets, macroeconomic co-movement, and financial integration across long time spans.
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 Fourier Johansen cointegration when you are testing long-run equilibrium relationships among two or more I(1) variables and suspect that the cointegrating vector or its deterministic component has shifted gradually over the sample — for example due to gradual policy reform, demographic trends, or evolving market integration. It is the preferred choice over standard Johansen when residual diagnostics or prior knowledge suggest smooth drift in the long-run relationship, but you cannot confidently pin down a break date. Avoid it when breaks are genuinely abrupt (use Zivot-Andrews or Johansen with dummies instead), when the sample is very short (fewer than ~60 observations per variable), or when the series are not clearly I(1), as spurious cointegration results are still possible.
Strengths & limitations
- Detects cointegration that standard Johansen misses when the relationship experiences smooth, gradual structural change.
- Does not require the researcher to pre-specify a break date or break type — the Fourier terms are estimated from the data.
- Extends naturally to multiple cointegrating vectors (rank > 1), preserving the full power of the multivariate Johansen framework.
- Low-frequency Fourier terms avoid overfitting; a single sine-cosine pair often captures the dominant smooth shift.
- Applicable to common empirical settings such as energy markets, macroeconomic co-movement, and financial integration across long time spans.
- Critical values differ from standard Johansen tables and depend on the chosen Fourier frequency; incorrect frequency selection inflates size or reduces power.
- Requires a reasonably long time series — short samples make frequency estimation unreliable and reduce power against the alternative of cointegration.
- The Fourier approximation captures only smooth, periodic-like breaks; sudden discrete jumps may not be well-approximated and could still distort inference.
- Software implementation is less widespread than standard Johansen, requiring custom code (R, GAUSS, or Stata scripts) in most applied settings.
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 Fourier Johansen differ from standard Johansen cointegration?
Standard Johansen assumes the deterministic component (constant or trend) is fixed throughout the sample. Fourier Johansen adds sine and cosine terms to approximate gradual shifts in that component, so the long-run equilibrium is allowed to evolve smoothly without pre-specifying break dates.
How do I choose the Fourier frequency k?
Estimate the Fourier-augmented VECM for k = 1, 2, 3, … (typically up to 5) and select the k that minimises the sum of squared residuals or an information criterion such as AIC or BIC. Using a single low-frequency harmonic (k = 1 or 2) is usually sufficient and guards against overfitting.
Can I still use trace and max-eigenvalue statistics?
Yes. Both statistics are computed in the usual way from the canonical correlations of the augmented VECM. However, because the Fourier terms alter the asymptotic null distribution, you must use response-surface or simulation-based critical values calibrated to the Fourier-Johansen framework, not the standard Johansen tables.
What if I believe the break is abrupt rather than gradual?
If the structural change is a sudden discrete shift, the Fourier approximation is less appropriate. Consider instead the Johansen test with regime dummy variables, or the structural-break Johansen test that explicitly estimates a sharp break date.
Does this test work with small samples?
Power is markedly lower in small samples (fewer than ~60 observations per variable) because both the Fourier frequency selection and the canonical correlation estimation become imprecise. In such cases standard Johansen or simpler single-equation tests with known break dates are preferable.
Sources
- 1.Enders, W., & Lee, J. (2012). A unit root test using a Fourier series to approximate smooth breaks. Oxford Bulletin of Economics and Statistics, 74(4), 574–599.
- 2.Johansen, S. (1988). Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control, 12(2–3), 231–254.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Fourier Johansen cointegration. ScholarGate. https://scholargate.app/econometrics/fourier-johansen-cointegration