I am new to time series and am trying to forecast a data series in r; which has weekly data. I have a few questions related to the same:

  1. While trying to use auto.arima() model, it shows the optimum order is c(0,1,1) with drift, however, it does not mention anything about the seasonality where in my data there is seasonality in one week per year. Data series looks like thisenter image description here

Should I use any other model to catch the seasonality factor and include it in model?

  1. Also, while forecasting the same, i get this result enter image description here

but while, trying to check accuracy i get error as "Error in window.default(x, ...) : 'start' cannot be after 'end'"

  1. I also tried Holt-Winters which gave the following result on my dataset: enter image description here

used this function: aust=window(train_1,start=2016) fit1=hw(aust,seasonal = "additive") autoplot(fit1)

Please help me understand how should i incorporate seasonal component in the model? is my Auto.arima model better than Holt-Winters? which one should i go with for forecasting ? and how to check the forecasting accuracy ? Many Thanks !

  • 1
    $\begingroup$ Its Holt-Winters, not Halt's Winter... $\endgroup$ – Richard Hardy Jun 13 at 8:54
  • $\begingroup$ your approach assumes that seasonality is stochastic as compared to deterministic ( think 51 dummies rather than 1 seasonal ar coefficient) . Both arise normally with weekly data and one needs to determine which approach is best for your data. Why don't you post your data in a csv file if possible. $\endgroup$ – IrishStat Jun 13 at 11:47

First of all try seasonal arima model like: arima(data,c(1,0,0), c(1,0,0) this is an example of 1 At model with 1AR seasonality...... To get better understanding of seasonality you have to see acf pacf graphs along with some decomposition techniques to look for seasonality.


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.