Am dealing with wind data experiencing some heavy tailed characteristics.
@Thomas Walther, thanks for the response. Sorry for the mix up in the question.
Yes it's a univariate time series. I've used fractional-ARIMA to forecast, but it tends to converge towards the mean. This forecast drastically improves when wavelet transformation is applied to the data, then forecast with f-arima. Attached is the plot.
@Paul R. Yarnold, Thank you so much, I'll go through the journal you've shared. I'm really interested in getting more forecasting methods that can give improved prediction accuracy.
As is generally the case, use theory to get a model and fit it. Fitting
a stationary time series will always produce after long intervals a
movement toward zero.
Herman
@Thomas Walther and @Herman Rubin, Thank you for your response. I'll consider it.
Dear Dennis.
I will recommend you to use exponential smoothing in package forecast. There is function ETS which can help you improve you prediction. But you have not to apply this function as it is in package forecast in R. You have to do some optimizations in order to find the good parameters wich fit well your data.
- Badges
- Science topic