04-13-2012 04:38 PM
I have a large data set that I'm using for my modeling. My data needs to be forecasted to Dec'2012. I have to end my data at Dec'11 so I can forecast 12 months out. The problem is that some series ( I have 20), stop short of the Dec'11 timeline, so I extrapolate the data to Dec'11 though proc forecast. When I do that since the price is on a down-trend the data comes out negative, which is impossible for the type of data. What is the norm for preparing time series data so that it doesn't show up negative but still need that observation to the end period of the time (eg. Dec'11).. I used PROC FORECAST, METHOD=WINTERS..