Fourier GLS (Fourier Generalized Least Squares)
Also known as: Fourier GLS, Fourier-based GLS, Fourier flexible GLS, spectral GLS
Fourier GLS embeds low-frequency trigonometric (Fourier) terms into a generalized least squares framework to capture smooth, gradual structural change in a time series without requiring the researcher to specify when or how many breaks occurred. The approach is particularly valued in unit root testing and cointegration analysis where conventional break-date assumptions may be arbitrary.
Key highlights
- Captures smooth, gradual structural change without requiring the researcher to pre-specify break dates or the number of breaks.
- Nests the standard GLS model as a special case when Fourier amplitudes are zero, enabling a natural specification test.
- Retains the efficiency advantage of GLS over OLS when errors are serially correlated or heteroscedastic.
- Applicable to unit root testing, cointegration, and general regression settings in a unified framework.
- Parsimonious: a single low-frequency sine-cosine pair can approximate a wide variety of smooth shifts.
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 GLS when you suspect smooth, continuous structural change in the deterministic component of a time series but cannot identify precise break dates — for example, gradually shifting inflation targets, evolving monetary policy stances, or slow demographic trends. It is well suited to unit root and stationarity testing when conventional break-point tests over-reject due to misspecified break dates. Avoid it when structural change is abrupt rather than smooth, when the sample is very short (fewer than ~50 observations), or when there is strong reason to model explicit regime switches — in those cases threshold or Markov-switching models are more appropriate.
Strengths & limitations
- Captures smooth, gradual structural change without requiring the researcher to pre-specify break dates or the number of breaks.
- Nests the standard GLS model as a special case when Fourier amplitudes are zero, enabling a natural specification test.
- Retains the efficiency advantage of GLS over OLS when errors are serially correlated or heteroscedastic.
- Applicable to unit root testing, cointegration, and general regression settings in a unified framework.
- Parsimonious: a single low-frequency sine-cosine pair can approximate a wide variety of smooth shifts.
- Cannot capture sharp, discontinuous breaks; if the true break is abrupt, the trigonometric approximation fits poorly and power falls.
- Frequency selection adds a pre-testing step that is not fully accounted for in the subsequent inference, potentially distorting test size in finite samples.
- Critical values for unit root tests must be simulated; standard ADF tables are not applicable.
- Performance deteriorates in very short samples where the Fourier terms consume many degrees of freedom.
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 Fourier GLS different from standard GLS?
Standard GLS corrects only for serial correlation or heteroscedasticity in the error term. Fourier GLS additionally augments the deterministic part of the model with sine and cosine terms to absorb unknown smooth structural change, making it robust to misspecification of the trend or break dates.
How do I choose the Fourier frequency k?
Grid-search over k = 1, 2, ..., floor(T/2) and select the value that minimises the sum of squared residuals (or an information criterion such as AIC). In practice, k = 1 or 2 suffices for most macroeconomic applications; larger values risk over-fitting.
Can Fourier GLS detect multiple structural breaks?
Yes, indirectly. A single low-frequency Fourier component can approximate two or more smooth shifts. However, if breaks are sharp and numerous, explicit multiple-break tests (e.g., Bai-Perron) are more powerful.
What critical values should I use for Fourier GLS unit root tests?
Standard ADF or PP critical values are incorrect here. Use the response-surface or simulation-based critical values tabulated in Enders and Lee (2012) or generate them by Monte Carlo for your specific sample size and model specification.
Is Fourier GLS suitable for panel data?
Extensions to panel unit root testing with Fourier terms exist (e.g., Nazlioglu and Karul, 2017), but the method was originally designed for univariate time series. For panel applications, consult the dedicated panel Fourier unit root literature.
Sources
- 1.Becker, R., Enders, W., & Hurn, S. (2004). A general test for time dependence in parameters. Journal of Applied Econometrics, 19(7), 899-906.
- 2.Enders, W., & Lee, J. (2012). The flexible Fourier form and Dickey-Fuller type unit root tests. Economics Letters, 117(1), 196-199.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Fourier GLS. ScholarGate. https://scholargate.app/econometrics/fourier-gls