3
$\begingroup$

I have a time series with muliple seasonal cycles, which are 24 and 168 hours for my case. I would like to use Double Seasonal Exponential Smoothing method to forecast, which was published by James W. Taylor (http://users.ox.ac.uk/~mast0315/). The link for his article is here

When I check Taylor's paper and all other papers in the literature, I saw that AR(1) adjustment has been implemented to the residuals. Is it a coincidence to make AR(1) correction to residuals of all the time series, that have been used in these papers?

I am using "forecast" package in R by Rob Hyndman, http://cran.r-project.org/web/packages/forecast/forecast.pdf

Firstly, I run the code without using an AR(1) model for the errors.

    data <- scan("data.dat", skip=1)
    datatimeseries <- ts(data)
    library(forecast)
    dataforecasts <- dshw(datatimeseries, 24, 168, armethod=FALSE)
    res=dataforecasts$residuals
    acf(res)
    pacf(res)

Below, you can see the ACF and PACF graphs of the residuals. enter image description here enter image description here

As you can see there are some serious non-zera autocorrelation. I can not do AR(1) correction as papers suggest because PACF does not show it.I do not know what to do. I am sorry if it's a silly question however I am so new to time series. Would you please care to explain what to do next step-by-step and elaborately. Thanks in advance

$\endgroup$
1
$\begingroup$

The AR(1) error would help a lot. But if you want something more general and flexible, use the tbats function instead.

$\endgroup$
  • $\begingroup$ Mr. Hyndman, When I do AR(1), no matter what the graphs depict, I still get non-zero autocorrelations. That means I did not get a perfect model. thus, how I use this tbats function you mentioned and why? $\endgroup$ – ARAT Jun 26 '13 at 8:05
  • $\begingroup$ First, you may never get a perfect model. Second, I've already explained why -- it is more general and flexible. Third, if you want to know how to use a function, read the help file. $\endgroup$ – Rob Hyndman Jun 26 '13 at 13:30

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.