topic predict demand given price and intervention using PROC NLMIXED in Statistical Procedures

predict demand given price and intervention using PROC NLMIXED

csetzkorn — Wed, 16 Aug 2017 18:12:37 GMT

I have daily time series product data where the dependent variable is demand for a product in units and the indepdent variables are product price and a binary intervention dummy variable to indicate promotion. I have been told that I can use PROC NLMIXED for this (as I do not have SAS/ETS available). Just curious, has someone applied PROC NLMIXED for such a scenario and if so what nonlinear function would you suggest to start with and how do I determine appropriate seeding values. Any feedback would be very much appreciated!

Re: predict demand given price and intervention using PROC NLMIXED

Ksharp — Thu, 17 Aug 2017 13:39:47 GMT

Yes. It is called repeated measure mixed model.

you can use it to predict demand.

proc mixed;

class ......;

model demand=price ..............;

random int /subject=person;

repeated time /subject=person type=ar(1) ;

run;

Re: predict demand given price and intervention using PROC NLMIXED

csetzkorn — Thu, 17 Aug 2017 15:39:57 GMT

Sorry is this intended that you do not use nlmixed? Also would person be productid in my case? I have one dependent var units. At least 3 independent vars: date, price, intervention flag (binary dummy).

Re: predict demand given price and intervention using PROC NLMIXED

Ksharp — Fri, 18 Aug 2017 12:51:26 GMT

proc nlmixed fit the nonlinear mixed model,
while proc mixed fit the linear mixed model.
I don't have much experience with proc nlmixed.
but I think proc mixed is able to get job done.

"Also would person be productid in my case? I have one dependent var units. At least 3 independent vars: date, price, intervention flag (binary dummy)."
Yes. person means a strata. you can use productid replace person.
if you want predicted value ,use OUTPUT statement.



proc mixed;
class flag date;
model demand=price flag;
random int /subject=person;
repeated date/subject=person type=ar(1) ;
OUTPUT out=want pred=p ;
run;