As per my job requirement I have to do forecasting using only Holt Winter technique in R.I have weekly data for 2 and half years & I have to predict weekly.I'm planning to build time series with frequency 52. I'm seeing that HoltWinters() function in R can do only Additive trend-Additive Seasonal (A,A) & Additive trend -Multiplicative Seasonal Model (A,M). I want to try other models like (M,A),(M,M). I saw these models can be done by ets() but the ets() has maximum frequency restriction of 24.

Can you please suggest me some method in R where I can use these models with 52 weeks of frequency?

Thanks in advance

  • $\begingroup$ Isn't this just a repetition of the thread at stats.stackexchange.com/questions/76322? $\endgroup$ – whuber Jun 17 '14 at 20:03
  • $\begingroup$ @whuber-In previous thread I was looking for forecasting with Holt-Winters technique without 104 weeks of data.Here I'm looking to add some other modeling approach for seasonality & trend in HoltWinters() function with 104 weeks of data.Hope I'm able to clarify your doubt. $\endgroup$ – R Learner Jun 18 '14 at 12:06
  • $\begingroup$ Despite those technical differences, both questions seem to concern a misconception about frequency and so they have received essentially identical answers. $\endgroup$ – whuber Jun 18 '14 at 13:37
  • $\begingroup$ Can you please help me to clear the misconception? $\endgroup$ – R Learner Jun 20 '14 at 6:02

Holt-Winters methods are inappropriate when you have frequency of 52 because they involve estimating 51 seasonal parameters which is clearly too many. With high frequency data you are much better off using Fourier terms. The analogous model to Holt-Winters is called a TBATS model, which you can fit using the tbats() function from the forecast package in R. See http://robjhyndman.com/hyndsight/forecasting-weekly-data/ for an example.

| cite | improve this answer | |

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.