Fourier WLS (Fourier Flexible Weighted Least Squares)
Also known as: Fourier WLS, Fourier-weighted least squares, smooth break WLS, Fourier flexible regression
Fourier WLS is a time-series regression technique that embeds low-frequency Fourier trigonometric terms into a Weighted Least Squares framework to capture smooth, gradual structural breaks in means or trends without requiring the researcher to pre-specify their location, timing, or number.
Key highlights
- Detects smooth, gradual structural breaks without requiring pre-specification of break dates or number of breaks.
- The low-frequency Fourier approximation avoids overfitting: a single frequency term captures the dominant structural shift parsimoniously.
- Combining Fourier terms with WLS yields efficient estimates even under heteroscedastic errors.
- Flexible enough to approximate a wide class of smooth nonlinear time trends through the trigonometric basis.
- Nested specification: if structural change is absent, the Fourier terms are insignificant and the model collapses to standard WLS.
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 WLS when you have a moderately long time series (roughly T ≥ 60) whose mean or trend appears to shift gradually rather than abruptly, and when you cannot or do not wish to pre-specify break dates. It is well suited to macroeconomic aggregates, exchange rates, and commodity prices that exhibit slow regime changes. Do not use it for series with sharp, discontinuous structural breaks — sharp breaks are better handled by Bai-Perron or Chow tests. Also avoid it when the sample is very short (T < 40), as the Fourier approximation will be poorly identified.
Strengths & limitations
- Detects smooth, gradual structural breaks without requiring pre-specification of break dates or number of breaks.
- The low-frequency Fourier approximation avoids overfitting: a single frequency term captures the dominant structural shift parsimoniously.
- Combining Fourier terms with WLS yields efficient estimates even under heteroscedastic errors.
- Flexible enough to approximate a wide class of smooth nonlinear time trends through the trigonometric basis.
- Nested specification: if structural change is absent, the Fourier terms are insignificant and the model collapses to standard WLS.
- Poorly suited to abrupt, discrete structural breaks — trigonometric terms spread the break over the whole sample and may distort estimates.
- Optimal frequency selection via grid search is somewhat data-driven; different selection criteria (AIC vs. BIC) can yield different chosen frequencies.
- Requires a reasonably long time series (T ≥ 60) for stable estimation; short samples inflate the risk of spurious Fourier patterns.
- The approach is primarily univariate or single-equation; extending it to multivariate systems (e.g., Fourier-VECM) adds considerable complexity.
- Interpretation of the Fourier coefficients themselves is not economically intuitive — they characterize the shape of the break, not its magnitude at a specific date.
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 many Fourier frequencies should I include?
Enders & Lee (2012) recommend searching over k = 1, 2, 3, 4, 5 and selecting the single k that minimises the sum of squared residuals or an information criterion. Including more than one frequency risks overfitting and complicates inference.
What is the difference between Fourier WLS and a standard Fourier unit-root test?
Fourier unit-root tests (e.g., Enders-Lee FADF) add Fourier terms to a Dickey-Fuller regression to test for a unit root robustly against smooth breaks. Fourier WLS is a general estimation framework that uses Fourier augmentation plus observation weights to model and estimate a stationary or trend-stationary series efficiently under heteroscedasticity.
Can Fourier WLS be used in panel data?
The core idea extends to panel settings, but the implementation becomes considerably more complex — the optimal frequency may differ across cross-sections, and pooled vs. heterogeneous estimation must be chosen carefully. Stand-alone panel Fourier estimators exist in the literature but are not a routine extension.
What if my break is sharp rather than smooth?
If a Chow test or visual inspection reveals an abrupt jump, Fourier WLS is inappropriate. Instead, use Bai-Perron multiple structural break tests or include dummy variables at the identified break date.
How do I choose the weights in the WLS step?
A common practice is to run an auxiliary OLS regression of squared residuals on the regressors (or a function of time) to estimate the variance function, then use the inverse of fitted variances as weights. FGLS-style iteration can improve efficiency when heteroscedasticity is severe.
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.Gallant, A. R. (1984). The Fourier flexible form. American Journal of Agricultural Economics, 66(2), 204–208.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Fourier WLS. ScholarGate. https://scholargate.app/econometrics/fourier-wls