How can state space models be applied to non-stationary time series data?
Instruction: Explain the concept of state space models and discuss how they can be used to model and forecast non-stationary time series data.
Updated
Example Answer
A state space model separates the observations from an underlying state and describes how that state evolves. The state might include a changing level, slope, or seasonal component; the observation equation describes how those components produce the measured value.
For demand with a drifting baseline, I could use a local-level or local-trend model instead of forcing the levels to be stationary. In a linear Gaussian formulation, a Kalman filter updates the state estimate as observations arrive. That is not the same as automatically refitting every model parameter. I would assess how quickly the state can change, how observation and process noise are estimated, and how the model behaves across gaps or breaks. Flexibility still needs validation at the forecast horizon.
Make it your own
Define the state components and which changes you expect the model to track.
Why this works
It explains where nonstationarity enters the model.
Follow-up to practice
What controls how quickly the estimated level adapts?