Robust SARIMA Model
Also known as: robust SARIMA, outlier-resistant SARIMA, robust seasonal ARIMA, M-estimator SARIMA
Robust SARIMA extends the classical Seasonal ARIMA framework by replacing the standard least-squares criterion with a robust loss function — such as an M-estimator — so that outliers and heavy-tailed innovations in seasonal time series cannot distort parameter estimates or invalidate forecasts.
Key highlights
- Resists distortion from additive outliers, innovative outliers, and level shifts without requiring manual data cleaning.
- Delivers reliable seasonal and trend decomposition even in contaminated series.
- Parameter estimates and forecasts remain close to those you would obtain from the clean data.
- Applicable to the full range of SARIMA orders, making it a drop-in robustified alternative to the classical model.
- Formal M-estimation foundation provides asymptotic theory and valid inference under mild contamination.
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 Robust SARIMA when you have a seasonal time series that likely contains outliers, measurement errors, or occasional structural shocks — common in macroeconomic indicators, energy demand, and retail sales data. It is especially valuable when forecasting accuracy matters more than a single anomaly, or when standard SARIMA diagnostics reveal non-normal, heavy-tailed residuals. Do not use it when the series is clean and Gaussian: the robustness comes at a modest efficiency cost, and classical SARIMA is then slightly more efficient. Also avoid it when you need a quick, interpretable baseline; the added complexity of the robust iteration requires more careful software handling.
Strengths & limitations
- Resists distortion from additive outliers, innovative outliers, and level shifts without requiring manual data cleaning.
- Delivers reliable seasonal and trend decomposition even in contaminated series.
- Parameter estimates and forecasts remain close to those you would obtain from the clean data.
- Applicable to the full range of SARIMA orders, making it a drop-in robustified alternative to the classical model.
- Formal M-estimation foundation provides asymptotic theory and valid inference under mild contamination.
- Estimation is iterative and computationally heavier than classical SARIMA, which can matter for high-frequency long series.
- Implementation requires specialised software (e.g., R packages robets, robustarima, or custom M-estimator code); not available in standard econometric suites out of the box.
- Modest efficiency loss relative to classical SARIMA when the data are actually Gaussian and outlier-free.
- Model selection (order identification via ACF/PACF) is still based on the structure of the data and can be complicated by contamination before robust estimation is applied.
- Distinguishing additive from innovational outliers in practice requires diagnostic care and can affect the cleaning strategy.
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 does Robust SARIMA differ from simply removing outliers before fitting standard SARIMA?
Manual removal requires knowing which observations are outliers in advance and risks deleting real signal. Robust SARIMA down-weights potential outliers automatically during estimation, using formal influence functions, so the threshold is data-driven and the procedure is part of the model fit rather than a pre-processing judgement call.
Which ψ function should I choose — Huber or Tukey bisquare?
Huber's function provides a good balance: it is quadratic for small residuals (like OLS) and linear for large ones, ensuring continuity. Tukey's bisquare redescends to zero for very large residuals, offering stronger resistance to extreme outliers but at the cost of a non-convex objective. For most economic series, Huber is a safe default; use bisquare when outliers are severe.
Can I use standard AIC/BIC for order selection in Robust SARIMA?
Standard information criteria based on Gaussian likelihood are distorted by outliers. Prefer robust model-selection criteria (e.g., robust AIC based on M-estimator log-likelihood) or use the ACF/PACF of robust residuals after a preliminary contamination-resistant fit.
Does Robust SARIMA handle all four types of outliers?
Most implementations directly address additive outliers (AO) and innovational outliers (IO). Level shifts (LS) and temporary changes (TC) can also be incorporated by augmenting the model with dummy regressors once the outlier type is identified, following the approach popularised in the X-12-ARIMA literature.
Is Robust SARIMA the same as SARIMA with robust standard errors?
No. Robust standard errors (HAC, sandwich estimators) correct the covariance matrix of coefficient estimates for heteroscedasticity or autocorrelation but do not change how the coefficients themselves are estimated. Robust SARIMA changes the objective function so that outlying observations have less influence on the coefficient estimates themselves.
Sources
- 1.Muler, N., Peña, D., & Yohai, V. J. (2009). Robust estimation for ARMA models. The Annals of Statistics, 37(2), 816–840.
- 2.Franses, P. H., & Ghijsels, H. (1999). Additive outliers, GARCH and forecasting volatility. International Journal of Forecasting, 15(1), 1–9.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Robust SARIMA model. ScholarGate. https://scholargate.app/econometrics/robust-sarima-model