1
$\begingroup$

This is probably a pretty basic question but I'd like to understand how you choose a seasonality number for the first forecast period in a Holt-Winters model.

If you need to forecast 8 months ahead would you begin with the seasonality number that corresponds to the same month 1 year ago? Example: use April '15's seasonality number to forecast April ''16's?

$\endgroup$
1
$\begingroup$

Yes.

In Holt-Winters, you will initialize and update three different components, one for the level, one for the trend and one for the seasonality. The seasonality component actually has as many entries as there are periods in one seasonal cycle - 4 for quarterly data, 12 for monthly data etc.

When you forecast out, the forecast includes the most recent level component estimate, the most recent trend component estimate (extrapolated, possibly dampened) and the most recent seasonality component estimate for the seasonal period you are interested in. Thus, when forecasting for next April, you'd use the entry corresponding to April.

I recommend this section on Holt-Winters in Hyndman's and Athanasopoulos' excellent free online forecasting textbook. Note in particular which seasonal entry is used in forecasting: $s_{t-m+h_m^+}$, where $s$ contains the updated seasonal components, $t$ is the forecast time period, $m$ is the length of the seasonal cycle, and

$$ h_m^+ = \lfloor (h-1)\ \text{mod}\ m\rfloor +1$$

for a forecast horizon $h$ ensures that you take the correct seasonal entry.

$\endgroup$
  • $\begingroup$ You are welcome. If my answer helped you, please consider upvoting and/or accepting it. Thank you! $\endgroup$ – Stephan Kolassa May 11 '16 at 18:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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