Hi, that’s a good question :) Before recording this I, actually, ran multiple models in SAS, and I also used R to run an autoarima function to see its recommendation. This model was recommended by R. As the model was complex enough, I used it as an example in this tutorial to show how to build a complex ARIMA model in SAS. Given the present autocorrelations, I would definitely have to continue modelling, though :) I would run ARIMA(1,1,1)(1,1,0)12 and if that wouldn’t work, I’d combine both seasonal AR and MA processes. ArIMA processes are not straightforward and require a lot of trial and error before you can find something that perfectly models your data and allows you to forecast future periods more or less accurately.
and as for seasonal, do i really need to do differencing for 1,12 because it seems i have to delete my 12 data. Because my data have seasonal as well but my lecturer says 1 differencing is enough.
That’s a good question. It all depends on how much data you can afford to lose. If you have over 10 years of data, losing one year shouldn’t be a big deal. Also, how strong is the seasonality? In reality, if you have a trend, differencing at lag1 should make your data stationary then AR and MA processes can easily take care of seasonality. The simpler the model, the better. In the tutorial I was aiming at showing the most complex cases and how to realize them in SAS. I hope it helps :)
Hello, I just watched few of your videos on arima modeling. Would you be willing to help mi figure out the parameters of my models? It woud help mi a lot because I am not greate at english.
@@sastutorials5245 So, I use arima modelling to predict development of parameters of some mortality models. I took advise from your comment section and used auto.arima in R to identify parameters of arima model for my time series a then tried to apply it in the SAS EG. But in case of ARIMA (2,2,0) or basically any model with AR (2) I am not able to get same result in SAS as in R (I mean completely different). I don’t know if I am doing something wrong or is this possible?
@@Jaja0812 Oh, I see. I know it may look confusing when you are trying to generate the same model in different software and get different results. But it’s ok! They’re using slightly different algorithms so don’t worry about that. Your results will never be the same :)
@@sastutorials5245 and do you consider results of auto.arima to be the best arima model or do you just use it to have some starting point?.. By the way, thank you very much for your time.
@@Jaja0812 My pleasure :) I would say that auto.arima is great, and you may go with its results, but I like checking a few models against the one I consider the best for comparison.
Thank you for your efforts! It is very helpful.
Kai Chang I am happy you found it helpful!
hai may i know why dont you include 1,12 in AR model at 44:49 ?
Hi, that’s a good question :) Before recording this I, actually, ran multiple models in SAS, and I also used R to run an autoarima function to see its recommendation. This model was recommended by R. As the model was complex enough, I used it as an example in this tutorial to show how to build a complex ARIMA model in SAS.
Given the present autocorrelations, I would definitely have to continue modelling, though :) I would run ARIMA(1,1,1)(1,1,0)12 and if that wouldn’t work, I’d combine both seasonal AR and MA processes. ArIMA processes are not straightforward and require a lot of trial and error before you can find something that perfectly models your data and allows you to forecast future periods more or less accurately.
and as for seasonal, do i really need to do differencing for 1,12 because it seems i have to delete my 12 data. Because my data have seasonal as well but my lecturer says 1 differencing is enough.
That’s a good question. It all depends on how much data you can afford to lose. If you have over 10 years of data, losing one year shouldn’t be a big deal. Also, how strong is the seasonality? In reality, if you have a trend, differencing at lag1 should make your data stationary then AR and MA processes can easily take care of seasonality. The simpler the model, the better. In the tutorial I was aiming at showing the most complex cases and how to realize them in SAS. I hope it helps :)
Hello, I just watched few of your videos on arima modeling. Would you be willing to help mi figure out the parameters of my models? It woud help mi a lot because I am not greate at english.
Hi, thank you for your comment. I think I’ll be able to direct you, for sure. So, what is your question in particular :)
@@sastutorials5245 So, I use arima modelling to predict development of parameters of some mortality models. I took advise from your comment section and used auto.arima in R to identify parameters of arima model for my time series a then tried to apply it in the SAS EG. But in case of ARIMA (2,2,0) or basically any model with AR (2) I am not able to get same result in SAS as in R (I mean completely different). I don’t know if I am doing something wrong or is this possible?
@@Jaja0812 Oh, I see. I know it may look confusing when you are trying to generate the same model in different software and get different results. But it’s ok! They’re using slightly different algorithms so don’t worry about that. Your results will never be the same :)
@@sastutorials5245 and do you consider results of auto.arima to be the best arima model or do you just use it to have some starting point?.. By the way, thank you very much for your time.
@@Jaja0812 My pleasure :) I would say that auto.arima is great, and you may go with its results, but I like checking a few models against the one I consider the best for comparison.