Loading tsb…
Combine a local level, trend, or Fourier seasonality in a state-space model. Filter observations, smooth the historical states, and forecast with prediction intervals.
A missing observation skips the measurement update; smoothing uses later observations to refine earlier estimates.
A local linear trend has a level and slope state. Forecast intervals include state and observation uncertainty.
Add Fourier components to a trend, then estimate noise variances by maximum likelihood.