Fourier VAR Model
Fourier-Augmented Vector Autoregression Model · Also known as: Fourier VAR, smooth structural break VAR, trigonometric VAR, Fourier-augmented VAR
The Fourier VAR model extends the standard Vector Autoregression by replacing fixed deterministic terms with Fourier trigonometric components, allowing the intercept (and optionally the trend) to shift gradually and smoothly over time. This eliminates the need to pre-specify the number, timing, or shape of structural breaks in a multivariate time-series system.
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 Fourier VAR when you are modelling a multivariate time-series system and suspect gradual, smooth structural change in the mean or trend — for example, after a long economic transition, a slow policy shift, or a secular trend change — but cannot pinpoint exact break dates. It is particularly valuable when the standard VAR produces serially correlated residuals or when unit-root pre-tests are sensitive to the chosen deterministic specification. Avoid it when the true breaks are sharp (use dummy-variable or threshold VAR instead), when the sample is short (fewer than roughly 50 time periods, as Fourier terms consume degrees of freedom), or when the number of endogenous variables is large relative to T (curse of dimensionality applies equally here).
Strengths & limitations
- Accommodates smooth, gradual structural change without requiring knowledge of break dates or break number.
- Reduces misspecification of the deterministic component, leading to better-behaved residuals and more reliable inference on VAR coefficients.
- Subsumes the standard VAR as a special case (when Fourier terms are insignificant) and is easy to test against it with an F-test.
- Compatible with standard VAR extensions: impulse responses, variance decompositions, and Granger causality tests can all be computed after estimation.
- Robust to over-fitting from unnecessary structural break dummy variables.
- Consumes additional degrees of freedom for each Fourier frequency included, which is costly in small samples or large systems.
- Can only approximate smooth, periodic-type breaks; sharp regime switches require threshold or Markov-switching VAR.
- Frequency selection is data-driven and may vary across bootstrap replications, adding a pre-testing uncertainty that is rarely accounted for.
- Interpretation of Fourier-augmented deterministic terms is less intuitive than a simple breakpoint dummy for practitioners.
Frequently asked
How is the Fourier VAR different from a VAR with structural break dummies?
Dummy-variable VARs assume breaks are sharp and require the researcher to specify dates in advance. The Fourier VAR instead uses smooth trigonometric terms that approximate gradual change, requiring only the choice of a frequency parameter selected from the data. It is more appropriate when break timing is unknown and when change is continuous rather than abrupt.
How do I choose the optimal Fourier frequency k?
Grid-search over integer values of k from 1 to a small upper bound (often 5). For each candidate k, estimate the augmented VAR and record AIC, BIC, or residual sum of squares. The k that minimises the chosen criterion is retained. A frequency of k = 1 implies one full smooth cycle over the sample, while higher k implies faster oscillations.
Can I still perform Granger causality tests after adding Fourier terms?
Yes. Once the Fourier-augmented VAR is estimated and the residuals pass autocorrelation and normality checks, standard Wald or F-tests for Granger causality are conducted on the lag coefficient blocks exactly as in a standard VAR. The Fourier terms are part of the deterministic specification and do not change the logic of the causality test.
Does the Fourier VAR require stationary data?
Like the standard VAR, it is applied after appropriate differencing if variables are integrated. The Fourier augmentation addresses smooth breaks in the deterministic component; it does not substitute for first-differencing or VECM specification when the system is cointegrated.
What sample size is adequate for a Fourier VAR?
As a rough guideline, at least 50 time observations are needed for a small bivariate system with one Fourier frequency. Each additional variable and each additional Fourier frequency multiplies the parameter count, so larger systems require proportionally longer series to maintain adequate degrees of freedom.
Sources
- 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. DOI: 10.1111/j.1468-0084.2011.00662.x ↗
- 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. DOI: 10.1111/j.1467-9892.2006.00478.x ↗
How to cite this page
ScholarGate. (2026, June 3). Fourier-Augmented Vector Autoregression Model. ScholarGate. https://scholargate.app/en/econometrics/fourier-var-model
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.
- Fourier ARDL Bounds TestEconometrics↔ compare
- Fourier Granger CausalityEconometrics↔ compare
- Fourier VECMEconometrics↔ compare
- Structural Break VAR ModelEconometrics↔ compare
- Structural VAREconometrics↔ compare
- Vector AutoregressionEconometrics↔ compare