Help on transfering parameter estimation to math expression

[TonyF]

I am first time SAS forcast studio user, and wondering if anyone in the community can help me understand how to transfer the parameter eatimation of a model to a math expression( the actuall formula).

Here is the estimation I got from SAS:

Component	Parameter	Estimate	Standard Error	t Value	Approx Pr > |t|
CS	MA1_1	0.32743	0.06042	5.42	<.0001
CS	MA1_2	0.08164	0.06408	1.27	0.2038
CS	MA1_3	0.28965	0.0608	4.76	<.0001
JP	SCALE	0.71904	0.01027	69.98	<.0001
JP	DEN1_1	0.03581	0.01478	2.42	0.0161
JP	DEN1_2	0.03784	0.01463	2.59	0.0103

CS is my dependent variable and JP is my independent variable.

Thanks!

[ballardw]

As a minimum to come close to answering this question we would basically need the code used to generate your estimates as the answer depends on which procedure, the model statement and any options.

[TonyF]

Hi Ballardw

Thanks for the reply. It is an ARIMA model.

here is the SAS code.

Proc HPFARIMASPEC

  * Model: SUBSETARIMA

  * Label: CS =  D=(1) Q=( 1 2 3 ) NOINT  + JP :  D=(1) DEN=( 1 2 )

  */

MODELREPOSITORY = work.temp

SPECNAME=LEAF_1

SPECLABEL="ARIMA:  CS  ~ D = (1)  Q = 3   NOINT  +  INPUT: Dif(1) JP  DEN = 2"

SPECTYPE=SUBSETARIMA

SPECSOURCE=FSUI

;

FORECAST SYMBOL = CS TRANSFORM = NONE

NOINT

DIF = ( 1 )

Q = ( 1 2 3 ) ;

INPUT SYMBOL = JP

TRANSFORM = NONE

DIF = ( 1 ) 

DEN = ( 1 2 );

ESTIMATE

METHOD=CLS

CONVERGE=0.0010

MAXITER=50

DELTA=0.0010

SINGULAR=1.0E-7  ;

run;

[rselukar]

Hello Tony,

I am not sure what you mean by an expression (or a math formula).  The estimated model is an ARIMA transfer function model.  These models are discussed in many time series analysis books, e.g. a book by Box and Jenkins, also see the ARIMA doc (and its references): SAS/ETS(R) 13.2 User's Guide

In any event, the specified model can be described as follows:

Let JP1 denote the first difference of JP, e_t denote the error at time t, and B denote the backward shift operator.  The model is:

CS_t = CS_(t-1) + [ 0.71904 /(1 - 0.03581B - 0.03784B^2)] JP1_t + e_t - 0.32743 e_(t-1) - 0.08164 e_(t-2) - 0.28965 e_(t-2)

This is not a formula for computing the forecasts of CS.  It is just a description of the model with estimated parameters.

Rajesh

[TonyF]

Hi rselukar,

This is excatly what I need! Thanks!

[TonyF]

Is it possible for me to set a formula for my independent varaibale (JP in this case)? So basically telling SAS what is the probability of JP going to each different value base on its probability distribution. Eg, if JP is 1000, then 20% it will be 500, and 80% it will be 2000 in next time point. And ask SAS forcast studio to run the simulation to predict the outcome in a certain period?

[rselukar]

It is possible.  You are describing a what-if analysis.  In order to do this you must create a score function based on this model.  See the SCORE statement documentation and example in the HPFENGINE procedure.

[udo_sas]

Hello -

Alternatively you should be able to use the scenario analysis feature in SAS Forecast Studio.

Thanks,

Udo

[TonyF]

Hi Udo,

Can you give me some details of how to do it in scenario analysis? From what I know, I can manully put in some value for my independent variable, and SAS will produce the new predicted value base on the input. However, it is different in my case.

I know how much,and what is the probability JP_t will go up or down if I know the CS_t-1. The probability of JP going upward or downward is actually base on the previous CS value, if CS is big, then the probability of JP going downward increase, and vice versa. I have the probabilty formula for JP already.

Ultimately, I want to run a simulation to predict how many time my JP will go up or down in a certain period in the future.

Is it that possible in scenatio analysis?

Thanks

[udo_sas]

Hello -

This is not possible using SAS Forecast Studio currently.

Thanks,

Udo

[TonyF]

Hi Udo,

Is there any other SAS software can do that?

We have base, enterprise guide, miner, and forecast studio so far.

Thanks

[udo_sas]

Hello -

You may be able to build some econometric model to do what you have in mind.

As part of SAS Forecast Server you will have access to SAS/ETS.

Adding for his thoughts.

Thanks,

Udo

[TonyF]

Hi rselukar,

I am wondering do you have a link or can you tell me where I can find the documentation for the SCORE statement you mentioned?

Thanks

[udo_sas]

See Chapter 22 "Using Forecasting Model Score Files and DATA Step Functions" of the SAS® Forecast Server Procedures User’s Guide

[TonyF]

Thanks! Will check it out