This book is in Open Review. I want your feedback to make the book better for you and other readers. To add your annotation, select some text and then click the on the pop-up menu. To see the annotations of others, click the button in the upper right hand corner of the page

6.2 Recursive relation

Similarly to how it was done for the pure additive model in Section 5.2, we can show what the recursive relation will look like for the pure multiplicative one (the logic here is the same, the main difference is in working with logarithms instead of the original values): \[\begin{equation} \begin{aligned} \log y_{t+h} = & \mathbf{w}_{m_1}' \mathbf{F}_{m_1}^{\lceil\frac{h}{m_1}\rceil-1} \log \mathbf{v}_{t} + \mathbf{w}_{m_1}' \sum_{j=1}^{\lceil\frac{h}{m_1}\rceil-1} \mathbf{F}_{m_1}^{j-1} \log \left(\mathbf{1}_k + \mathbf{g}_{m_1} \epsilon_{t+m_1\lceil\frac{h}{m_1}\rceil-j}\right) + \\ & \mathbf{w}_{m_2}' \mathbf{F}_{m_2}^{\lceil\frac{h}{m_2}\rceil-1} \log \mathbf{v}_{t} + \mathbf{w}_{m_2}' \sum_{j=1}^{\lceil\frac{h}{m_2}\rceil-1} \mathbf{F}_{m_2}^{j-1} \log \left(\mathbf{1}_k + \mathbf{g}_{m_2} \epsilon_{t+m_2\lceil\frac{h}{m_2}\rceil-j}\right) + \\ & \dots \\ & \mathbf{w}_{m_d}' \mathbf{F}_{m_d}^{\lceil\frac{h}{m_d}\rceil-1} \log \mathbf{v}_{t} + \mathbf{w}_{m_d}' \sum_{j=1}^{\lceil\frac{h}{m_d}\rceil-1} \mathbf{F}_{m_d}^{j-1} \log \left(\mathbf{1}_k + \mathbf{g}_{m_d} \epsilon_{t+m_d\lceil\frac{h}{m_d}\rceil-j}\right) + \\ & \log \left(1 + \epsilon_{t+h}\right) \end{aligned}. \tag{6.7} \end{equation}\] In order to see how this recursion works, we can take the example of ETS(M,N,N), for which \(m_1=1\) and all the other frequencies are equal to zero: \[\begin{equation} y_{t+h} = \exp\left(\mathbf{w}_{1}' \mathbf{F}_{1}^{h-1} \log\mathbf{v}_{t} + \mathbf{w}_{1}' \sum_{j=1}^{h-1} \mathbf{F}_{1}^{j-1} \log \left(\mathbf{1}_k + \mathbf{g}_{1} \epsilon_{t+h-j}\right) +\log \left(1 + \epsilon_{t+h}\right)\right) , \tag{6.8} \end{equation}\] or after inserting \(\mathbf{w}_{1}=1\), \(\mathbf{F}_{1}=1\), \(\mathbf{v}_{t}=l_t\), \(\mathbf{g}_{1}=\alpha\) and \(\mathbf{1}_k=1\): \[\begin{equation} y_{t+h} = l_t \prod_{j=1}^{h-1} \left(1 + \alpha \epsilon_{t+h-j}\right) \left(1 + \epsilon_{t+h}\right) . \tag{6.9} \end{equation}\]

This recursion is useful in order to understand how the states evolve over time and in case of ETS(M,N,N) it allows obtaining the conditional expectation and variance. But in general for models with trend and / or seasonality, it cannot be used for the calculation of moments, as the one for the pure additive ADAM ETS. This is discussed in Section 6.3).