$$\begin{align*} \hat{y}_{t+h|t} &= (l_{t} \circ_{b} (b_{t} \circ_{d} \phi )) \circ_{s} s_{t+h-m(k+1)} \\ l_{t} &= \alpha (y_{t} \ominus_{s} s_{t-m}) + (1-\alpha)(l_{t-1} \circ_{b} (b_{t-1} \circ_{d} \phi ))\\ b_{t} &= \frac{\beta}{\alpha}(l_{t} \ominus_{b} l_{t-1}) + (1-\frac{\beta}{\alpha})b_{t-1} \\ s_{t} &= \gamma (y_{t} \ominus_{s}(l_{t-1} \circ_{b}(b_{t-1} \circ_{d} \phi))) +(1-\gamma)s_{t-m}\\ \end{align*}$$
  • $\alpha, \beta, \gamma, \phi$ : smoothing parameters. In particular, $\ \phi_{h} = \phi + \phi^2 + \cdots + \phi^{h}$ means the damping parameter.
  • $k$ : the quotient of $\frac{h-1}{m}$




Exponential Smoothing

N,N : Simple exponential smoothing

$$\begin{align*} y_{t+h|t} &= l_{t} \\ l_{t} &= \alpha y_{t} + (1-\alpha) l_{t-1} \\ \end{align*}$$

N,A : Simple exponential smoothing with additive seasonal

$$\begin{align*} y_{t+h|t} &= l_{t} + s_{t+h-m(k+1)}\\ l_{t} &= \alpha (y_{t} - s_{t-m}) +(1-\alpha)l_{t-1} \\ s_{t} &= \gamma(y_{t} - l_{t-1}) +(1-\gamma)s_{t-m}\\ \end{align*}$$

N,M : Simple exponential smoothing with multiplicative seasonal

$$\begin{align*} y_{t+h|t} &= l_{t} s_{t+h-m(k+1)}\\ l_{t} &= \alpha (y_{t} / s_{t-m}) +(1-\alpha)l_{t-1} \\ s_{t} &= \gamma(y_{t} / l_{t-1}) +(1-\gamma)s_{t-m}\\ \end{align*}$$

A,N : Holt’s linear method

$$\begin{align*} y_{t+h|t} &= l_{t} + hb_{t}\\ l_{t} &= \alpha y_{t} +(1-\alpha)(l_{t-1} + b_{t-1})\\ b_{t} &= \beta^{*}(l_{t} - l_{t-1}) +(1-\beta^{*})b_{t-1}\\ \end{align*}$$

A,A : Additive Holt-Winters’ method

$$\begin{align*} y_{t+h|t} &= l_{t} + hb_{t} + s_{t+h-m(k+1)}\\ l_{t} &= \alpha (y_{t} - s_{t-m}) +(1-\alpha)(l_{t-1} + b_{t-1})\\ b_{t} &= \beta^{*}(l_{t} - l_{t-1}) +(1-\beta^{*})b_{t-1}\\ s_{t} &= \gamma(y_{t} - l_{t-1} - b_{t-1}) +(1-\gamma)s_{t-m}\\ \end{align*}$$

A,M : Multiplicative Holt-Winters’ method

$$\begin{align*} y_{t+h|t} &= ( l_{t} + hb_{t} ) s_{t+h-m(k+1)}\\ l_{t} &= \alpha (y_{t} / s_{t-m}) +(1-\alpha)(l_{t-1} + b_{t-1})\\ b_{t} &= \beta^{*}(l_{t} - l_{t-1}) +(1-\beta^{*})b_{t-1}\\ s_{t} &= \gamma(y_{t} / (l_{t-1} - b_{t-1})) +(1-\gamma)s_{t-m}\\ \end{align*}$$

Ad,N : Additive damped trend method

$$\begin{align*} y_{t+h|t} &= l_{t} + \phi_{h}b_{t}\\ l_{t} &= \alpha y_{t} +(1-\alpha)(l_{t-1} + \phi b_{t-1})\\ b_{t} &= \beta^{*}(l_{t} - l_{t-1}) +(1-\beta^{*})\phi b_{t-1}\\ \end{align*}$$

Ad,A : Additive damped trend method with additive seasonal

$$\begin{align*} y_{t+h|t} &= l_{t} + \phi_{h}b_{t} + s_{t+h-m(k+1)}\\ l_{t} &= \alpha (y_{t} - s_{t-m}) +(1-\alpha)(l_{t-1} + \phi b_{t-1})\\ b_{t} &= \beta^{*}(l_{t} - l_{t-1}) +(1-\beta^{*})\phi b_{t-1}\\ s_{t} &= \gamma(y_{t} - l_{t-1} - \phi b_{t-1}) +(1-\gamma)s_{t-m}\\ \end{align*}$$

Ad,M : Holt-Winters’ damped method

$$\begin{align*} y_{t+h|t} &= ( l_{t} + \phi_{h}b_{t} ) s_{t+h-m(k+1)}\\ l_{t} &= \alpha (y_{t} / s_{t-m}) +(1-\alpha)(l_{t-1} + \phi b_{t-1})\\ b_{t} &= \beta^{*}(l_{t} - l_{t-1}) +(1-\beta^{*})\phi b_{t-1}\\ s_{t} &= \gamma(y_{t} / (l_{t-1} - \phi b_{t-1})) +(1-\gamma)s_{t-m}\\ \end{align*}$$




ETS Models

Additive models

$$e_{t} = y_{t} - \hat{y}_{t|t-1}$$

ETS(A,N,N)

$$\begin{align*} y_{t} &= l_{t-1} + \epsilon_{t} \\ l_{t} &= l_{t-1} + \alpha \epsilon_{t} \\ \end{align*}$$

ETS(A,N,A)

$$\begin{align*} y_{t} &= l_{t-1} + s_{t-m} + \epsilon_{t} \\ l_{t} &= l_{t-1} + \alpha \epsilon_{t} \\ s_{t} &= s_{t-m} + \gamma \epsilon_{t} \\ \end{align*}$$

ETS(A,N,M)

$$\begin{align*} y_{t} &= l_{t-1} s_{t-m} + \epsilon_{t} \\ l_{t} &= l_{t-1} + \alpha \epsilon_{t}/s_{t-m} \\ s_{t} &= s_{t-m} + \gamma \epsilon_{t}/l_{t-1} \\ \end{align*}$$

ETS(A,A,N)

$$\begin{align*} y_{t} &= l_{t-1} + b_{t-1} + \epsilon_{t} \\ l_{t} &= l_{t-1} + b_{t-1} + \alpha \epsilon_{t} \\ b_{t} &= b_{t-1} + \beta \epsilon_{t} \\ \end{align*}$$

ETS(A,A,A)

$$\begin{align*} y_{t} &= l_{t-1} + b_{t-1} + s_{t-m} + \epsilon_{t} \\ l_{t} &= l_{t-1} + b_{t-1} + \alpha \epsilon_{t} \\ b_{t} &= b_{t-1} + \beta \epsilon_{t} \\ s_{t} &= s_{t-m} + \gamma \epsilon_{t} \\ \end{align*}$$

ETS(A,A,M)

$$\begin{align*} y_{t} &= (l_{t-1} + b_{t-1}) s_{t-m} + \epsilon_{t} \\ l_{t} &= l_{t-1} + b_{t-1} + \alpha \epsilon_{t}/s_{t-m} \\ b_{t} &= b_{t-1} + \beta \epsilon_{t}/s_{t-m} \\ s_{t} &= s_{t-m} + \gamma \epsilon_{t}/(l_{t-1} + b_{t-1}) \\ \end{align*}$$

ETS(A,Ad,N)

$$\begin{align*} y_{t} &= l_{t-1} + \phi b_{t-1} + \epsilon_{t} \\ l_{t} &= l_{t-1} + \phi b_{t-1} + \alpha \epsilon_{t} \\ b_{t} &= \phi b_{t-1} + \beta \epsilon_{t} \\ \end{align*}$$

ETS(A,Ad,A)

$$\begin{align*} y_{t} &= l_{t-1} + \phi b_{t-1} + s_{t-m} + \epsilon_{t} \\ l_{t} &= l_{t-1} + \phi b_{t-1} + \alpha \epsilon_{t} \\ b_{t} &= \phi b_{t-1} + \beta \epsilon_{t} \\ s_{t} &= s_{t-m} + \gamma \epsilon_{t} \\ \end{align*}$$

ETS(A,Ad,M)

$$\begin{align*} y_{t} &= (l_{t-1} + \phi b_{t-1}) s_{t-m} + \epsilon_{t} \\ l_{t} &= l_{t-1} + \phi b_{t-1} + \alpha \epsilon_{t}/s_{t-m} \\ b_{t} &= \phi b_{t-1} + \beta \epsilon_{t}/s_{t-m} \\ s_{t} &= s_{t-m} + \gamma \epsilon_{t}/(l_{t-1} + \phi b_{t-1}) \\ \end{align*}$$

Multiplicative models

$$e_{t} = \frac{y_{t}}{\hat{y}_{t|t-1}} - 1$$

ETS(M,N,N)

$$\begin{align*} y_{t} &= l_{t-1}(1 + \epsilon_{t}) \\ l_{t} &= l_{t-1}(1 + \alpha \epsilon_{t}) \\ \end{align*}$$

ETS(M,N,A)

$$\begin{align*} y_{t} &= (l_{t-1} + s_{t-m})(1 + \epsilon_{t}) \\ l_{t} &= l_{t-1} + \alpha (l_{t-1} + s_{t-m}) \epsilon_{t} \\ s_{t} &= s_{t-m} + \gamma (l_{t-1} + s_{t-m}) \epsilon_{t} \\ \end{align*}$$

ETS(M,N,M)

$$\begin{align*} y_{t} &= l_{t-1} s_{t-m}(1 + \epsilon_{t}) \\ l_{t} &= l_{t-1} (1 + \alpha \epsilon_{t}) \\ s_{t} &= s_{t-m} (1 + \gamma \epsilon_{t}) \\ \end{align*}$$

ETS(M,A,N)

$$\begin{align*} y_{t} &= (l_{t-1} + b_{t-1})(1 + \epsilon_{t}) \\ l_{t} &= (l_{t-1} + b_{t-1})(1 + \alpha \epsilon_{t}) \\ b_{t} &= b_{t-1} + \beta (l_{t-1} + b_{t-1}) \epsilon_{t} \\ \end{align*}$$

ETS(M,A,A)

$$\begin{align*} y_{t} &= (l_{t-1} + b_{t-1} + s_{t-m})(1 + \epsilon_{t}) \\ l_{t} &= l_{t-1} + b_{t-1} + \alpha (l_{t-1} + b_{t-1} + s_{t-m}) \epsilon_{t} \\ b_{t} &= b_{t-1} + \beta (l_{t-1} + b_{t-1} + s_{t-m}) \epsilon_{t} \\ s_{t} &= s_{t-m} + \gamma (l_{t-1} + b_{t-1} + s_{t-m}) \epsilon_{t} \\ \end{align*}$$

ETS(M,A,M)

$$\begin{align*} y_{t} &= (l_{t-1} + b_{t-1}) s_{t-m} (1 + \epsilon_{t}) \\ l_{t} &= (l_{t-1} + b_{t-1})(1 + \alpha \epsilon_{t}) \\ b_{t} &= b_{t-1} + \beta (l_{t-1} + b_{t-1}) \epsilon_{t} \\ s_{t} &= s_{t-m}( 1 + \gamma \epsilon_{t}) \\ \end{align*}$$

ETS(M,Ad,N)

$$\begin{align*} y_{t} &= (l_{t-1} + \phi b_{t-1})(1 + \epsilon_{t}) \\ l_{t} &= (l_{t-1} + \phi b_{t-1})(1 + \alpha \epsilon_{t}) \\ b_{t} &= \phi b_{t-1} + \beta (l_{t-1} + \phi b_{t-1}) \epsilon_{t} \\ \end{align*}$$

ETS(M,Ad,A)

$$\begin{align*} y_{t} &= (l_{t-1} + \phi b_{t-1} + s_{t-m})(1 + \epsilon_{t}) \\ l_{t} &= l_{t-1} + \phi b_{t-1} + \alpha (l_{t-1} + \phi b_{t-1} + s_{t-m}) \epsilon_{t} \\ b_{t} &= \phi b_{t-1} + \beta (l_{t-1} + \phi b_{t-1} + s_{t-m}) \epsilon_{t} \\ s_{t} &= s_{t-m} + \gamma (l_{t-1} + \phi b_{t-1} + s_{t-m}) \epsilon_{t} \\ \end{align*}$$

ETS(M,Ad,M)

$$\begin{align*} y_{t} &= (l_{t-1} + \phi b_{t-1}) s_{t-m} (1 + \epsilon_{t}) \\ l_{t} &= (l_{t-1} + \phi b_{t-1})(1 + \alpha \epsilon_{t}) \\ b_{t} &= \phi b_{t-1} + \beta (l_{t-1} + \phi b_{t-1}) \epsilon_{t} \\ s_{t} &= s_{t-m}( 1 + \gamma \epsilon_{t}) \\ \end{align*}$$
 

 

 


Reference
  1. Forecasting: Principles and Practice (3rd ed)
  2. Statsmodels - ETSModel
Related Articles