Fourier ARIMA Model
Fourier-Augmented Autoregressive Integrated Moving Average Model · Also known as: Fourier ARIMA, ARIMA with Fourier terms, trigonometric ARIMA, Fourier-flexible ARIMA
The Fourier ARIMA model augments a standard ARIMA specification with trigonometric sine and cosine terms, allowing it to capture smooth, gradual structural change and flexible nonlinear seasonality without specifying the exact timing or number of breaks in advance. It is widely used in applied macroeconometrics and finance for series exhibiting slowly evolving dynamics.
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 Fourier ARIMA when you have a univariate economic or financial time series that shows gradual, smooth structural change or complex seasonal patterns that do not fit a simple seasonal ARIMA or a single-break model. It is particularly appropriate when the number, timing, and nature of breaks are unknown, as the flexible Fourier form avoids pre-testing bias. Do not use it when breaks are sharp and abrupt (the smooth Fourier approximation is then inefficient), when the series is very short (T < 50), or when multiple-equation dynamics require a VAR or VECM framework.
Strengths & limitations
- Captures gradual structural change and nonlinear seasonality without specifying break dates.
- Avoids pre-testing bias inherent in Chow-type break tests by treating change as a smooth process.
- Parsimonious: a small k (often 1 or 2 frequency pairs) is sufficient to approximate most smooth departures.
- Integrates naturally with standard ARIMA frameworks and can be estimated with conventional maximum likelihood software.
- Well-suited for long macroeconomic series with unknown or multiple regime shifts.
- Cannot adequately model sharp, sudden structural breaks; in such cases a segmented ARIMA or Markov-switching model dominates.
- Selection of k requires an additional model-selection step; poor choice inflates or deflates the effective degrees of freedom.
- Requires a sufficiently long series (T ≥ 50, ideally ≥ 100) for the Fourier approximation to be reliable.
- Interpretation of Fourier amplitudes is less intuitive than standard ARIMA coefficients.
- May overfit if k is not disciplined by an information criterion.
Frequently asked
How is the Fourier ARIMA different from SARIMA?
SARIMA handles fixed, periodic seasonality (e.g., exactly 12-month cycles in monthly data) through seasonal differencing and seasonal AR/MA terms. Fourier ARIMA uses sine and cosine waves to approximate any smooth, possibly non-integer or evolving seasonal or trend pattern, making it more flexible but also requiring frequency selection via an information criterion.
How do I choose k, the number of Fourier frequency pairs?
Fit the model for k = 1, 2, …, k_max (commonly k_max = 5) and select the k that minimises AIC or BIC. In practice k = 1 or 2 is often sufficient for macroeconomic series, and parsimony should guide the choice to avoid overfitting.
Can Fourier ARIMA identify when a structural break occurred?
No. The smooth Fourier approximation spreads structural change over the entire sample and cannot pinpoint a single break date. If localising a break is the goal, use a Chow test, Bai-Perron procedure, or a Markov-switching model instead.
Does the model require stationarity?
The ARIMA component handles non-stationarity through differencing (order d). You should determine d via standard unit root tests, then add Fourier terms to the differenced (or undifferenced, if d = 0) series. Applying Fourier terms to an undifferenced I(1) series without accounting for the stochastic trend is a common error.
What software can estimate a Fourier ARIMA?
R packages such as forecast (with xreg for Fourier regressors generated by fourier()) and tsfknn, as well as Python's statsmodels ARIMA with manually constructed Fourier columns, support this model. Eviews and Stata can also handle it by including sine/cosine variables as additional regressors in an ARIMA framework.
Sources
- Enders, W., & Lee, J. (2012). The flexible Fourier form and Dickey-Fuller type unit root tests. Economics Letters, 117(1), 196-202. DOI: 10.1016/j.econlet.2012.04.081 ↗
- Becker, R., Enders, W., & Hurn, S. (2004). A general test for time dependence in parameters. Journal of Applied Econometrics, 19(7), 899-906. DOI: 10.1002/jae.751 ↗
How to cite this page
ScholarGate. (2026, June 3). Fourier-Augmented Autoregressive Integrated Moving Average Model. ScholarGate. https://scholargate.app/en/econometrics/fourier-arima-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.
- ARIMA modelEconometrics↔ compare
- SARIMA modelEconometrics↔ compare