Fourier ADF Unit Root Test
Also known as: Fourier ADF test, FADF test, Flexible Fourier ADF, Fourier-based ADF unit root test
The Fourier ADF unit root test extends the standard Augmented Dickey-Fuller framework by incorporating low-frequency Fourier terms into the deterministic component. This allows the test to approximate smooth, gradual structural breaks in the level or trend of a time series without requiring prior knowledge of break number, timing, or form.
Key highlights
- Captures smooth, gradual structural breaks without requiring prior specification of their number, timing, or shape.
- Gains power over standard ADF and Zivot-Andrews when the true break is continuous rather than abrupt.
- Simple to implement: only a single additional selection step (optimal Fourier frequency) is needed beyond standard ADF.
- Flexible — low Fourier frequencies can approximate a wide variety of nonlinear deterministic processes.
- Avoids the pre-testing distortions that arise from first locating a break date and then conditioning inference on it.
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 Fourier ADF test when you suspect structural breaks in the level or trend of a time series are smooth and gradual rather than abrupt — for example, after monetary regime changes, energy transitions, or long-run demographic shifts. It is appropriate when the number or timing of breaks is unknown and you want to avoid the pre-testing bias of sharp-break tests. It is less suitable when breaks are truly sudden and one-off (Zivot-Andrews or Lee-Strazicich tests are preferable in that case), when the sample is very short (T < 50), or when multiple sharp breaks need to be located precisely.
Strengths & limitations
- Captures smooth, gradual structural breaks without requiring prior specification of their number, timing, or shape.
- Gains power over standard ADF and Zivot-Andrews when the true break is continuous rather than abrupt.
- Simple to implement: only a single additional selection step (optimal Fourier frequency) is needed beyond standard ADF.
- Flexible — low Fourier frequencies can approximate a wide variety of nonlinear deterministic processes.
- Avoids the pre-testing distortions that arise from first locating a break date and then conditioning inference on it.
- Requires simulation-based critical values that depend on the chosen frequency; standard ADF tables are inapplicable.
- May lack power against very abrupt, single-episode breaks compared to dedicated sharp-break tests.
- Choosing too high a Fourier frequency risks overfitting the deterministic component and under-rejecting the unit root null.
- Sample size requirements are non-trivial; performance deteriorates noticeably when T is below 50-80 observations.
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 is the Fourier ADF different from the Zivot-Andrews test?
Zivot-Andrews models a single sharp break at an unknown date, shifting the intercept or trend instantaneously. The Fourier ADF instead uses sine and cosine terms to capture smooth, gradual changes. If the true break is sudden, Zivot-Andrews is more powerful; if the break is gradual, the Fourier ADF has the advantage.
How do I choose the optimal Fourier frequency k?
Estimate the auxiliary regression (without the unit root restriction) for k = 1, 2, 3, 4, 5 and choose the k that minimises the sum of squared residuals or the AIC. Most applications find k = 1 or k = 2 sufficient for capturing gradual structural change.
Can I use standard ADF critical values for the Fourier ADF test?
No. The inclusion of Fourier terms changes the limiting distribution of the test statistic. You must use the simulation-based critical values tabulated in Enders and Lee (2012), which depend on the chosen frequency k and the deterministic specification (intercept only, or intercept and trend).
What if the Fourier terms are jointly insignificant?
If a joint F-test on all Fourier terms fails to reject the null of no Fourier component, the series likely does not contain a smooth gradual break, and standard ADF or PP tests are appropriate. Enders and Lee (2012) recommend this pre-test before relying on Fourier-adjusted critical values.
How many lags should I include in the Fourier ADF regression?
Select the number of augmenting lags p to eliminate residual autocorrelation, just as in standard ADF. AIC or BIC are the most common criteria, with an upper bound of roughly 12(T/100)^(1/4) as a practical guide.
Sources
- 1.Becker, R., Enders, W., & Lee, J. (2006). A stationarity test in the presence of an unknown number of smooth breaks. Journal of Time Series Analysis, 27(3), 381-409.
- 2.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.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Fourier ADF unit root test. ScholarGate. https://scholargate.app/econometrics/fourier-adf-unit-root-test