Macro by Mark 的 Cookie.我们使用必要 Cookie 处理登录、语言和安全。经您同意,我们还会将可选 Cookie 用于保存偏好、分析,以及通过 Meta(Facebook)Pixel 衡量广告效果。请阅读我们的Cookie 政策。
Error / trend / seasonality state-space framework that decomposes a series into smoothed components and forecasts forward.
Follow the same equations and assumptions one section at a time or in a continuous reading.
Hover or focus a symbol defined in this model to find its exact matches. Select it to keep the highlight; Escape clears the selection.
The ETS framework recasts every exponential smoothing method as a state-space model with a single source of error. The model is defined by three choices: the error type (A for additive, M for multiplicative), the trend type (N for none, A for additive, Ad for additive damped, M for multiplicative, Md for multiplicative damped), and the seasonal type (N for none, A for additive, M for multiplicative). These three choices create a grid of candidate models, of which about 15 are commonly used.
The single-source-of-error assumption means that one innovation drives both the observation and all state updates simultaneously. This is not the most general state-space formulation - a multiple-source-of-error model would have separate disturbances for the observation and each state. But the single-source assumption gives each model a closed-form innovation representation, which means the likelihood can be evaluated without the Kalman filter for additive error models. This is what makes ETS computationally fast.
Measurement equation for : the observation is the sum of the previous level, trend, and seasonal factor, plus an additive innovation.
Level transition: the level advances by the trend and absorbs a fraction of the innovation.
Trend transition: the trend persists and absorbs a fraction of the innovation.
Seasonal transition: the seasonal factor for this period within the cycle is updated by a fraction of the innovation.
With damping: the trend shrinks by factor each period. For , the trend decays geometrically toward zero over the forecast horizon.
Measurement equation for : multiplicative error and multiplicative seasonality. The seasonal factor scales the level+trend, and the error is a relative perturbation.
Conditioning on information at T separates the sample from the projection. The forecast can be smooth even when later realized data will not be.
The bridge point is shared so the eye can see the handoff. After that point, the red line is a model forecast, not observed history.
Do not score the forecast segment as realized data or read it as a policy promise.
Each future innovation contributes through its impulse weight . Unless the future is degenerate, longer horizons carry more conditional variance.
The band widens when shocks accumulate through the model recursion.
Do not treat the band edge as an alternative deterministic scenario.
The history segment is realized data available at the forecast origin.
The forecast path is a conditional expectation, so it can be smooth while future realizations remain volatile.
Band width is horizon-specific uncertainty, not a second deterministic transition.