04-27-2014 10:35 AM
I need some help to write scoring code from proc autoreg. What is the actual code that can replicate the predicted quantities from the output out = dataset?
It's easy enough to calculate the predicted model component (using standard coefficients) but how does one incorporate the AR coefficients?
Any help will be greatly appreciated.
I'm struggling even with a simple example: consider the example given in SAS online documentation:
ul = 0; ull = 0;
do time = -10 to 36;
u = + 1.3 * ul - .5 * ull + 2*rannor(12346);
y = 10 + .5 * time + u;
if time > 0 then output;
ull = ul; ul = u;
auto-reg model - produce an output dataset with predicted quatitites: QUESTION: HOW CAN I USE MODEL EFFECTS TO REPLICATE PREDICTED VALUES?
proc autoreg data=a outest = autoreg_parms;
aut: model y = time / nlag=2 method=ml;
output out = pred_a p = pred_y ;
04-27-2014 01:27 PM
There are two types of predicted values for AR models, conditional and unconditional, depending on your needs. Which one would you like to compute? See
04-28-2014 05:05 AM
Thanks for your reply. I'm after the conditional mean values. The descriptions in the user guide are all nice and well for t > nlag. However, there is nothing on how to start the predicted series: the auto-regressive formula
v_t = -phi_2*v_t-2 - phi_1*v_t-1 (for AR(2) say) doesn't say how to compute v_1 and v_2. This is my whole problem.
Do you know how SAS actually does this?