Obsidian | Level 7

## Time series model about AIC matrix

Hi everyone,

I know use minic statement we can get minimum information criterion and best p,q selection by BIC method and also a matrix consists of different combination of AR&MA like below(a part of scrrenshot).

PROC ARIMA DATA= Training;

IDENTIFY VAR = Y(1) MINIC;

RUN;

Quit;

The question is that can I get AIC matrix like above?? so I can compare AIC and BIC together? After comparison, I can choose a good selection of p, q and use the selected p,q to forecast loan price after 2018 to 2024.

And I know scan statement can also give us p,q selection, but how to interpret the selection result?

Note: I take d=1, but p+d=0? I think p+d should equal to at least 1, cause d=1. And p+d=11, which means this selection can be p=10, q=0, which arima model is (10,1,0).

Best,

Michelle

1 ACCEPTED SOLUTION

Accepted Solutions
SAS Employee

## Re: Time series model about AIC matrix

Hi Michelle,

PROC ARIMA only outputs the BIC when the MINIC option is specified and does not provide an option to generate a matrix of AIC values for the set of candidate models.  Box, Jenkins and Reinsel (1994) provide a good discussion of the algorithm used by the MINIC option in PROC ARIMA.

Regarding your second question, if you are specifying something like:

identify var=y(1) scan;

then the Tentative Order Selection Tests table identifies candidate models for the differenced series.  Any differencing orders that have already been specified in the VAR= option are not included in the (p+d) column of the Tentative Order Selection Tests table.  Based on the table you provided, the final models suggested by the combination of the SCAN option output and the differencing order you already specified in the VAR= option are:  ARIMA(0,1,1) and ARIMA(11,1,0).

Example 7.5 Using Diagnostics to Identify ARIMA Models in the PROC ARIMA documentation provides some additional insight into tentative order selection.  A link to that example is provided below:

http://go.documentation.sas.com/?docsetId=etsug&docsetTarget=etsug_arima_examples05.htm&docsetVersio...

I hope this helps!

DW

SAS Employee

## Re: Time series model about AIC matrix

Hi Michelle,

PROC ARIMA only outputs the BIC when the MINIC option is specified and does not provide an option to generate a matrix of AIC values for the set of candidate models.  Box, Jenkins and Reinsel (1994) provide a good discussion of the algorithm used by the MINIC option in PROC ARIMA.

Regarding your second question, if you are specifying something like:

identify var=y(1) scan;

then the Tentative Order Selection Tests table identifies candidate models for the differenced series.  Any differencing orders that have already been specified in the VAR= option are not included in the (p+d) column of the Tentative Order Selection Tests table.  Based on the table you provided, the final models suggested by the combination of the SCAN option output and the differencing order you already specified in the VAR= option are:  ARIMA(0,1,1) and ARIMA(11,1,0).

Example 7.5 Using Diagnostics to Identify ARIMA Models in the PROC ARIMA documentation provides some additional insight into tentative order selection.  A link to that example is provided below:

http://go.documentation.sas.com/?docsetId=etsug&docsetTarget=etsug_arima_examples05.htm&docsetVersio...

I hope this helps!

DW

Discussion stats