Singular Spectrum Analysis
Also known as: SSA, SVD-based decomposition
Singular Spectrum Analysis (SSA) is a nonparametric method for time-series decomposition and forecasting based on singular value decomposition (SVD) of a time-lagged embedding matrix. Introduced by Broomhead and King (1986) and developed further by Vautard, Yiou, and Ghil (1992), SSA decomposes time series into trend, oscillatory, and noise components without assuming any underlying model. It is particularly effective for short, noisy non-stationary signals where parametric approaches fail.
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
Apply SSA when you need to separate trend, oscillations, and noise in short, noisy time series without assuming a parametric model. It is ideal for climate data, economic indicators, seismic signals, and any domain with non-stationary behavior. SSA shines when signal-to-noise ratio is low or when the data has multiple overlapping components.
Strengths & limitations
- Nonparametric approach—does not require specification of model order or frequency components
- Works well with short time series and noisy data where classical methods struggle
- Provides simultaneous trend extraction and noise filtering
- Interpretable components with clear physical or business meaning
- Invertible and reversible—reconstructed components exactly sum to the original signal
- Requires careful choice of window length L—sensitivity analysis is needed to ensure robustness
- Grouping singular components into signal and noise is subjective; no automatic algorithm exists
- For very long time series, the embedding matrix becomes large, increasing memory and computational cost
- The method assumes stationarity of patterns within the window—unsuitable for highly non-stationary signals with abrupt regime changes
Frequently asked
How do I choose the optimal window length L?
L should be long enough to capture important patterns (typically L > 2 × dominant period) but short enough to have sufficient degrees of freedom (L < N/2). Use cumulative variance plots or cross-validation to assess sensitivity and guide the choice.
How do I separate signal from noise in the singular values?
The singular value spectrum often has a clear 'knee' where signal values drop off sharply and noise begins. Visual inspection of the scree plot combined with physical interpretation of eigenvectors helps identify the cutoff. More formal methods include MDL (minimum description length) and AIC.
Can SSA forecast into the future?
Yes, the Recurrent SSA (R-SSA) and Vector SSA (VSSA) variants can be used for forecasting. They use the reconstructed trend and oscillatory components to project forward. However, SSA itself is primarily a decomposition tool; forecasting requires additional modeling.
What is the relationship between SSA and PCA?
SSA is essentially PCA applied to a time-lagged embedding matrix. The difference is conceptual: PCA treats observations as independent, while SSA exploits the temporal structure of the data through the embedding matrix.
Can SSA handle multivariate time series?
Yes, multivariate SSA (MSSA) extends the method to handle several time series simultaneously. It constructs a block Hankel matrix with columns from multiple series and applies the same SVD and reconstruction steps.
Sources
- Broomhead, D. S., & King, G. P. (1986). Extracting qualitative dynamics from experimental data. Physica D: Nonlinear Phenomena, 20(2–3), 217–236. DOI: 10.1016/0167-2789(86)90031-X ↗
- Vautard, R., Yiou, P., & Ghil, M. (1992). Singular-spectrum analysis: A toolkit for short, noisy chaotic signals. Physica D: Nonlinear Phenomena, 58(1–4), 95–126. DOI: 10.1016/0167-2789(92)90103-T ↗
- Golyandina, N., Nekrutkin, V., & Zhigljavsky, A. (2001). Analysis of Time Series Structure: SSA and Related Techniques. Chapman and Hall/CRC. link ↗
How to cite this page
ScholarGate. (2026, June 3). Singular Spectrum Analysis. ScholarGate. https://scholargate.app/en/time-series/singular-spectrum-analysis
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.
- Independent Component AnalysisMachine learning↔ compare
- Kernel PCAMachine learning↔ compare
- Singular Value DecompositionNumerical Methods↔ compare