Quantcast
Channel: Statistics Help @ Talk Stats Forum - SAS
Viewing all articles
Browse latest Browse all 310

ARIMA and non-stationary ts

$
0
0
Hello everyone!
Can somebody suggests me, what you are doing in case that you have data table with 2 variables - date and for example number of something (500 days) and you want to predict and forecast this numerical variable in next maybe 200 days (until end of this year). the distribution is very not normal, data series is non stationary and with decreased trend and little bit seasonality (in weekends).

and if you try several arima (maybe about 30 try) - for example arima(1,0,1)...arima(5,0,5), different combinations of this p and q parameters and if no model gives well result, what should you do in that case?

for example, forecasted values are all in vertical decreasing or in vertical increasing, and I want to get similar as history data (with non stationary, zig zag curve and very little decrease trend..)

please tell me your experiences.

best

Viewing all articles
Browse latest Browse all 310

Trending Articles