The Holt-Winters additive method model is defined to be

\begin{align*} \hat{y}_{t+h|t} &= \ell_{t} + hb_{t} + s_{t+h-m(k+1)} \\ \ell_{t} &= \alpha(y_{t} - s_{t-m}) + (1 - \alpha)(\ell_{t-1} + b_{t-1})\\ b_{t} &= \beta^*(\ell_{t} - \ell_{t-1}) + (1 - \beta^*)b_{t-1}\\ s_{t} &= \gamma (y_{t}-\ell_{t-1}-b_{t-1}) + (1-\gamma)s_{t-m}, \end{align*}

but the text I am reading doesn't mention on the relevant page what precisely $h$ represents. I see that it scales $b_t$ and shifts $s_{t+h+m(k+1)}$, but it isn't obvious to me what the motivation or meaning of this value is. The passage mentions that $k$ is the integer part of some ratio $\frac{h-1}{m}$, but this did not sufficiently clarify what the quantity $h$ is. What is this $h$ representing?


Hyndman, R.J., & Athanasopoulos, G. (2018) Forecasting: principles and practice, 2nd edition, OTexts: Melbourne, Australia. OTexts.com/fpp3. Accessed on 2022-07-19.

Is $h$ the prediction horizon? If so, why does $h$ scale $b_t$ and shift $s_{t-m(k+1)}$?


1 Answer 1


$h$ is indeed the prediction horizon. $\hat{y}_{t+h|t}$ is the $h$-step ahead forecast made at time $t$, so it is for period $t+h$ and conditional on information up to period $t$; therefore the subscript is $t+h|t$.

Why does $h$ scale $b_t$? The term $b_t$ is the trend component, specifically its value at period $t$. If we forecast out $h$ periods, we expect $h$ increments due to the trend, the best estimate of which currently is $b_t$. Thus, the contribution of the trend to our point forecast is simply $hb_t$. (If the trend is dampened, you will see a dampening factor in this term, but this is an undampened trend model.)

Why does $h$ shift $s_{t+h-m(k+1)}$? We have a season of length $m$ periods. $k$ represents how many full seasonal cycles we forecast out. When we forecast out $h$ periods, it could be that we forecast less than one full seasonal cycle, e.g., six months in a yearly seasonal model. In this case, $h=6$ and $k=0$. Then $s_{t+h-m(k+1)}$ will "look back" to the last entry of the season component that corresponds to the period we are forecasting for. We have July; if we forecast out $h=6$ months, then $s_{t+h-m(k+1)}=s_{t+6-12(0+1)}=s_{t-6}$ will look back and pick the season component entry corresponding to January. If we forecast out more than one full cycle, $k>0$, and the index here will make sure we look at a component within the last cycle.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.